The challenge
Build a useful ML environment without internet access. Installation, updates and dependencies must be organised in advance.
Dutch critical-infrastructure organisation
MLOps without a connection outside.
A fully offline machine-learning platform, from GPU hardware to images, pipelines and model serving.
What was built
Kubeflow on RKE2, with a local image registry, private model storage and H100 GPU capacity. Required images and artefacts remain inside the environment.
The result
An in-house training and serving platform, with a managed route for controlled software and model imports.
VF / 01REFERENCE ARCHITECTURE
- People and applications: CHAT / API / DEVELOPER TOOLS
- Access · API gateway: IDENTITY / LIMITS / ROUTING
- AI agents, retrieval and models: TOOLS / SOURCES / MODELS
- GPU · Storage · MLOps: KUBERNETES / NVMe-oF / OBSERVABILITY
REFERENCE, NOT A NETWORK MAP
The pattern behind the solution.
A generic overview of the building blocks. Technical addresses, internal systems and client identities are excluded from publication.
Explore the architecture ↗